Zhejiang unveils Wuyin AI data foundation

In Zhejiang, a new data foundation for robots just materialized, pulling 50-plus firms to its launch.

The Zhejiang-based startup WUWENAI unveiled its first physical AI data foundation platform, dubbed Wuyin, at a Deqing venue that drew more than 50 robotics and AI players, including Horizon Robotics, Xingdong Jiyuan, Lingxin Dexterous Hand, Deep Robotics and Digu Robotics. Mandarin-language reporting indicates the event framed Wuyin as an industry-first “data foundation layer” designed to knit real-world data, high-fidelity simulation, and industrial deployment into a single closed loop.

Wuyin rests on three core capabilities. First is a high-quality data system that fuses real-world data collection with high-fidelity simulated datasets, signaling a deliberate push to bridge live operation and model training. Second is a high-value scenario ecosystem that rebuilds real applications in synthetic form, enabling rapid testing and iteration. Third is a Real2Sim2Real closed-loop toolchain that ties data acquisition, model training, scenario evaluation, and deployment into a single workflow. The platform’s architecture mirrors a broader policy-driven push in China to reduce development risk for robotics and embodied AI by tightly coupling data, simulation, and real-world deployment. Supply chain disclosures reveal a growing preference for end-to-end data platforms over standalone hardware or software stacks.

Wuyin’s scale is notable. The platform has already accumulated more than 1,000 terabytes of data and has open-sourced 10,000 hours of high-quality datasets, according to the launch materials. For data collection, Wuyin supports multiple modalities, including VR teleoperation, exoskeleton operation, UMI (unified motion capture) capture, first-person human data recording, and general motion capture. Cross-embodiment data transfer tools enable different robots to reuse the same dataset, a move that could compress development cycles for diverse robot platforms. In addition, the platform leverages generative simulation technologies to build a library of millions of “sim-ready” assets, mapping real-world environments into virtual simulations via a Real2Sim2Real workflow.

Current application scenarios highlighted at the launch include logistics and other industrial automation contexts, pointing to a pathway for domestic factories to accelerate embodied AI deployment without always building bespoke datasets from scratch. The event’s attendees, it’s noted, included both hardware-hardware players and software-centric AI firms, underscoring a convergence of capabilities that China’s regional ecosystems have been cultivating for years.

The momentum around Wuyin sits within a broader ecosystem of state- and private-backed capital flowing into embodied AI and robotics data infrastructure. In a parallel track, PsiBot, a Shanghai-based embodied AI company, raised around RMB 2 billion across angel and Pre-Series A rounds to speed up data collection and logistics deployment, with investors ranging from state-backed funds to strategic capital from large listed players. Mandarin-language reporting indicates PsiBot’s 21-degree-of-freedom exoskeleton data glove is central to its data-collection strategy, and the company has validated roughly 10,000 hours of data. The funding signals a broader incentive for cities and provinces to back data-centric robotics platforms that can scale across industry, particularly in logistics.

For practitioners watching China’s robot economy, Wuyin’s launch offers both a blueprint and a caution. On the upside, the Real2Sim2Real loop, plus a ready-made library of sim-ready assets and cross-robot data reuse, could dramatically shorten the path from pilot to production. On the downside, maintaining data quality at scale—and safeguarding IP and data governance across cross-robot datasets—will test enterprises as they move from proof-of-concept to large-scale deployment. In Zhejiang, these questions arrive at the factory floor with a concrete answer: a platform that treats data as a first-order product, not an afterthought.

Two concrete practitioner takeaways: first, the emphasis on cross-embodiment data reuse will pressure suppliers and operators to harmonize data standards and labeling across fleets, which can reduce duplication but raises data governance questions. second, the Real2Sim2Real workflow is a powerful accelerant, but firms must invest in high-fidelity simulators and asset libraries; the value hinges on matching the diversity of real-world scenarios to the simulated ones—and on keeping the asset library current as processes shift.

Sources

WUWENAI Launches Physical AI Data Infrastructure Platform “Wuyin” in Zhejiang

PsiBot Raises $280M to Accelerate Embodied AI Data Collection and Logistics Deployment